MMpred

MMpred applies distance-assisted multimodal conformation sampling to improve de novo protein structure prediction by enhancing conformation sampling efficiency and model accuracy.

Key Features:

• Multimodal Optimization Protocol: Employs a multimodal optimization protocol to improve the efficiency of conformation sampling and modeling accuracy.
• Modal Exploration (DMscore): Uses the structural similarity evaluation model DMscore to promote diversity among generated conformations and create populations across low-energy basins.
• Modal Maintaining (MNDcluster): Utilizes the adaptive clustering algorithm MNDcluster to organize populations and merge modalities by adjusting the annealing temperature to identify promising energy basins.
• Modal Exploitation: Applies a greedy search strategy to accelerate convergence within identified modalities.
• Distance Constraint Integration: Integrates distance constraints via a conformation scoring model to guide sampling and enhance structural diversity and accuracy.

Scientific Applications:

• Benchmarking on non-redundant proteins: Validated on a dataset of 320 non-redundant proteins, producing models with TM-score ≥ 0.5 for 268 cases and showing a 20.3% improvement over Rosetta when using the same distance constraints.
• Protein assembly modeling: Enables sampling of multiple promising energy basins with enhanced structural diversity, supporting improved model accuracy in protein assembly simulations.

Methodology:

Three-stage multimodal optimization comprising modal exploration using DMscore, modal maintaining with adaptive MNDcluster clustering and annealing temperature adjustment, modal exploitation via greedy search, and integration of distance constraints through a conformation scoring model.